pi-better-harness 0.7.0 → 0.8.0

package/node_modules/pi-better-background-tasks/README.md

@@ -99,9 +99,11 @@ After `/new`, `/resume`, fork, or switching to another session, the previous
 session's tasks keep running but are paused from Pi's side: watches do not poll,
 remote tmux output is not collected, and `timeout_seconds` deadlines are not
 enforced until that session is active again. An overdue deadline is enforced as
-soon as the session resumes, so a timeout can land late but is never skipped. A
-local process that exits in the meantime is recorded as finished, and its
-callback is delivered when its session resumes.
+soon as the session resumes, so a timeout can land late but is never skipped.
+While the same Pi process stays open, a local process that exits in the meantime
+is recorded as finished, and its callback is delivered when its session resumes.
+If you quit Pi first, nothing records that exit: when the session is resumed in a
+new Pi process, a task whose process is gone is marked lost.
 
 ## Watch conditions
 
@@ -151,7 +153,7 @@ evidence is reported as **observation incomplete**.
 Pass `operation_id` and `expected_exit_codes` on `bg_task_spawn`, `bg_task_watch`,
 or `bg_task` to declare intent before launch; a malformed declaration starts
 nothing. An exit code in `expected_exit_codes` (distinct integers 1-255, such as
-`[1]` for a no-match probe) is recorded as an **Expected failure** that needs no
+`[1]` for a no-match probe; a `0` is ignored) is recorded as an **Expected failure** that needs no
 action; signals and timeouts never are. When a task with an `operation_id`
 succeeds, earlier failed tasks with the same `operation_id`, kind, cwd, SSH
 target, and owner (the same session id, or for sessionless tasks the same Pi

package/node_modules/pi-better-background-tasks/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-better-background-tasks",
-  "version": "0.4.0",
+  "version": "0.5.0",
   "description": "Pi extension for durable background shell tasks, watchers, logs, and status inspection.",
   "license": "MIT",
   "type": "module",

package/node_modules/pi-better-background-tasks/src/failures.ts

@@ -1,7 +1,7 @@
 import { join } from "node:path";
 import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import {
-  activeFailures, failureAttentionHandled, failureIdentity, formatFailureSummary, markFailureAttentionDelivered,
+  actionableFailures, activeFailures, failureAttentionHandled, failureIdentity, formatFailureSummary, markFailureAttentionDelivered,
   observeFailures, pendingAttentionNote, pendingAttentionRows, pendingFailureAttention, readCommandIntent, readFailureState,
   type FailureEvent, type FailureState,
 } from "./shared-failure-observations.js";
@@ -12,6 +12,14 @@ import type { BackgroundTaskMeta } from "./types.js";
 
 export const failurePath = (id: string): string => join(taskDir(id), "failures.jsonl");
 export const failureSummary = (id: string): string => formatFailureSummary(readFailureState(failurePath(id)));
+/**
+ * The failure summary plus whether anything needs action. A navigator row leads with failure text
+ * only when something does; the quiet history line never displaces the command (#332).
+ */
+export function failureView(id: string): { text: string; actionable: boolean } {
+  const state = readFailureState(failurePath(id));
+  return { text: formatFailureSummary(state), actionable: actionableFailures(state).length > 0 };
+}
 
 export function recordFailure(meta: BackgroundTaskMeta, operation: string, summary: string, eventKey: unknown,
   options: { category?: string; expected?: boolean; incomplete?: boolean; evidence?: string; at?: number } = {}): void {

package/node_modules/pi-better-background-tasks/src/navigator-provider.ts

@@ -10,8 +10,8 @@ import {
 } from "./shared-navigator.ts";
 import { CustomEditor } from "@earendil-works/pi-coding-agent";
 import { Key, matchesKey, truncateToWidth } from "@earendil-works/pi-tui";
-import { activeFailures, failureLabel, readFailureState } from "./shared-failure-observations.js";
-import { failurePath, failureSummary } from "./failures.js";
+import { actionableFailures, failureLabel, readFailureState } from "./shared-failure-observations.js";
+import { failurePath, failureView } from "./failures.js";
 import { readLog } from "./logs.js";
 import { listMetasForOrigin, onMetaChanged, readMeta, writeMeta } from "./registry.js";
 import { stopTask } from "./runtime.js";
@@ -105,7 +105,9 @@ function isExpiredTerminalNavigatorRow(meta: BackgroundTaskMeta, now: number): b
 }
 
 function rowFromMeta(meta: BackgroundTaskMeta, now: number): BackgroundWorkRow {
-  const failure = failureSummary(meta.id);
+  // Only a failure that needs action replaces the row's command; history stays in the detail view (#332).
+  const view = failureView(meta.id);
+  const failure = view.actionable ? view.text : "";
   const elapsed = formatDuration((meta.endedAt ?? now) - meta.startedAt);
   return {
     providerId: "background-tasks",
@@ -129,7 +131,8 @@ function rowFromMeta(meta: BackgroundTaskMeta, now: number): BackgroundWorkRow {
 
 function detailFromMeta(meta: BackgroundTaskMeta | undefined, now: number, options?: { logTailLines?: number }): BackgroundWorkDetail | null {
   if (!meta) return null;
-  const failure = failureSummary(meta.id);
+  const view = failureView(meta.id);
+  const failure = view.text;
   const log = readLog(meta.logPath, options?.logTailLines ?? NAVIGATOR_DETAIL_ROWS);
   const command = commandLabel(meta);
   const metadata = [
@@ -164,7 +167,7 @@ function detailFromMeta(meta: BackgroundTaskMeta | undefined, now: number, optio
     title: meta.name || meta.id,
     status: meta.status,
     statusTone: toneForStatus(meta.status),
-    subtitle: failure ? failure.split("\n")[0]! : compactCommandLabel(meta),
+    subtitle: view.actionable ? failure.split("\n")[0]! : compactCommandLabel(meta),
     metadata,
     foldedSections: [{
       id: "command",
@@ -217,7 +220,7 @@ function secondaryLabel(meta: BackgroundTaskMeta): string | undefined {
 
 function factsForMeta(meta: BackgroundTaskMeta, now: number): string[] {
   const facts: string[] = [];
-  const incident = activeFailures(readFailureState(failurePath(meta.id)))[0];
+  const incident = actionableFailures(readFailureState(failurePath(meta.id)))[0];
   if (incident) facts.push(`${failureLabel(incident)}: ${incident.summary}`);
   if (meta.status === "running") {
     const stall = observeBackgroundTaskStall(meta, now);

package/node_modules/pi-better-background-tasks/src/output.ts

@@ -548,13 +548,16 @@ export function formatStatus(
   if (pageKind === "f") return formatLog(meta.id, { ...options, raw: true });
   if (pageKind === "t") return formatVerbose(meta, options);
   if (options.verbose) return formatVerbose(meta, options);
+  // A bg_task_list page cursor pages the list, not this task: say so rather than report it as a
+  // stale status cursor (#332).
+  const listCursor = pageKind === "l";
   const state = failureStateFor(meta.id);
   const resource = `status:${scopeKey(options)}:${meta.id}`;
   const revision = inspectStatusRevision({
     resource,
     contentRevision: contentRevision(meta),
     failureRevision: failureRevision(state),
-    cursor: options.cursor,
+    cursor: listCursor ? undefined : options.cursor,
   });
   if (options.cursor && revision.change === "none") {
     return assembleBackgroundContent({
@@ -574,6 +577,7 @@ export function formatStatus(
   // Change/reset and read-gap facts are short and decision-relevant: they are
   // budgeted with the decision section, ahead of long incident rows.
   const changeFacts = [
+    ...(listCursor ? ["cursor ignored: it is a bg_task_list page cursor; pass it to bg_task_list, or pass this status's statusCursor here."] : []),
     ...(revision.change === "failure" ? ["change=failure"] : []),
     ...(revision.reset ? [`reset=${revision.reset}`] : []),
     ...(log.error ? [`log unreadable: ${oneLine(log.error, 200)}; cannot treat this as an empty healthy log.`] : []),

package/node_modules/pi-better-background-tasks/src/runtime.ts

@@ -34,8 +34,11 @@ const remoteSessionStarts = new Map<string, Promise<CommandResult>>();
 const activePolls = new Set<string>();
 const logRetentionTimers = new Map<string, ReturnType<typeof setInterval>>();
 const LOG_RETENTION_CHECK_MS = 1000;
-const handoffTimers = new Map<string, ReturnType<typeof setInterval>>();
+const handoffTimers = new Map<string, ReturnType<typeof setTimeout>>();
+// The handoff check backs off from 250 ms to 4 s while the task keeps running,
+// and returns to 250 ms once its process is gone so the grace period is timed closely.
 const HANDOFF_CHECK_MS = 250;
+const HANDOFF_MAX_CHECK_MS = 4_000;
 const HANDOFF_LOST_GRACE_MS = 5_000;
 let scheduledWorkSuspended = false;
 const REMOTE_SESSION_POLL_MS = 100;
@@ -62,8 +65,9 @@ export const DEFAULT_WATCH_TIMEOUT_SECONDS = 15 * 60;
  * tasks keep running, but their watches do not poll, remote tmux output is not
  * collected, and their `timeout_seconds` deadlines are not enforced until that
  * session is active again; an overdue deadline is enforced immediately on resume.
- * A local process that exits meanwhile is recorded as terminal and its callback
- * is delivered when its session resumes.
+ * A local process that exits meanwhile, while this Pi process is still running,
+ * is recorded as terminal and its callback is delivered when its session resumes;
+ * after Pi quits, a resumed task whose process is gone is marked lost instead.
  */
 export function suspendScheduledWork(): void {
   scheduledWorkSuspended = true;
@@ -71,7 +75,7 @@ export function suspendScheduledWork(): void {
   for (const timer of remoteSessionTimers.values()) clearTimeout(timer);
   for (const timer of processTimeoutTimers.values()) clearTimeout(timer);
   for (const timer of logRetentionTimers.values()) clearInterval(timer);
-  for (const timer of handoffTimers.values()) clearInterval(timer);
+  for (const timer of handoffTimers.values()) clearTimeout(timer);
   handoffTimers.clear();
   watcherTimers.clear();
   remoteSessionTimers.clear();
@@ -528,7 +532,9 @@ function scheduleHandoff(
   stopHandoff(id);
   if (scheduledWorkSuspended) return;
   let deadSince: number | undefined;
-  const timer = setInterval(() => {
+  let delayMs = HANDOFF_CHECK_MS;
+  const check = () => {
+    handoffTimers.delete(id);
     const meta = readMeta(id);
     if (!meta) {
       stopHandoff(id);
@@ -554,16 +560,26 @@ function scheduleHandoff(
         clearProcessTimeout(id);
         stopLogRetention(id);
         markProcessLost(pi, meta, getActiveSession);
+        return;
       }
+      delayMs = HANDOFF_CHECK_MS;
+    } else {
+      delayMs = Math.min(HANDOFF_MAX_CHECK_MS, delayMs * 2);
     }
-  }, HANDOFF_CHECK_MS);
-  timer.unref();
-  handoffTimers.set(id, timer);
+    arm();
+  };
+  const arm = () => {
+    if (scheduledWorkSuspended) return;
+    const timer = setTimeout(check, delayMs);
+    timer.unref();
+    handoffTimers.set(id, timer);
+  };
+  arm();
 }
 
 function stopHandoff(id: string): void {
   const timer = handoffTimers.get(id);
-  if (timer) clearInterval(timer);
+  if (timer) clearTimeout(timer);
   handoffTimers.delete(id);
 }
 

package/node_modules/pi-better-background-tasks/src/shared-failure-observations.ts

@@ -104,24 +104,60 @@ export interface CommandIntent {
 export type CommandIntentField = keyof CommandIntent;
 const INTENT_FIELDS: readonly CommandIntentField[] = ["operationId", "attemptId", "expectedExitCodes"];
 /**
- * The one normalization of intent-bearing arguments: an explicit `null` intent field means
- * "not declared", exactly like an omitted one. Models routinely send optional fields as null, and
- * Pi's argument validation (0.87+) drops optional nulls before a tool runs, while the process log
- * and session record keep the raw arguments. Every reader of intent, the executing tool and the
- * replay alike, goes through this so both sides see the same declaration from the same input.
+ * Pi's own argument coercion for the intent fields (pi-ai `validateToolArguments` against the confined
+ * bash schema, where each field is `anyOf: [field schema, null]`). Pi applies it before execute, but the
+ * process log and session record keep the raw arguments; mirroring it lets the parent's replay read
+ * the declaration the child executed with (#332).
+ * - ids: a number or boolean becomes its string; `""` fails the pattern and falls through to null.
+ * - exit codes: array items that read as integers become integers (null -> 0, true -> 1);
+ *   a scalar `0`, `""`, or `false` falls through to null.
+ */
+function coerceInteger(value: unknown): unknown {
+  if (value === null) return 0;
+  if (typeof value === "string" && value.trim() !== "" && Number.isInteger(Number(value))) return Number(value);
+  if (typeof value === "boolean") return value ? 1 : 0;
+  return value;
+}
+function normalizeIntentField(key: CommandIntentField, value: unknown): unknown {
+  if (value == null) return undefined;
+  if (key === "expectedExitCodes") {
+    if (value === 0 || value === "" || value === false) return undefined;
+    if (!Array.isArray(value)) return value;
+    // 0 is a no-op declaration: exit 0 is already success. A list of only zeros declares nothing.
+    const codes = value.map(coerceInteger).filter((code) => code !== 0);
+    if (codes.length === 0 && value.length > 0) return undefined;
+    return codes.length === value.length && codes.every((code, index) => code === value[index]) ? value : codes;
+  }
+  if (value === "") return undefined;
+  return typeof value === "number" || typeof value === "boolean" ? String(value) : value;
+}
+/**
+ * The one normalization of intent-bearing arguments. An explicit `null` intent field means "not
+ * declared", exactly like an omitted one: models routinely send optional fields as null, and Pi's
+ * argument validation drops or nulls them before a tool runs, while the process log and session
+ * record keep the raw arguments. Values Pi would coerce are coerced the same way, and `0` is dropped
+ * from `expectedExitCodes`. Every reader of intent, the executing tool and the replay alike, and the
+ * exact operation identity go through this, so all see the same declaration from the same input.
+ * Malformed values are kept as they are for the validator to refuse.
  */
 export function withoutAbsentIntent<T>(args: T): T {
   if (!args || typeof args !== "object" || Array.isArray(args)) return args;
   const input = args as Record<string, unknown>;
-  if (!INTENT_FIELDS.some((key) => key in input && input[key] == null)) return args;
-  const out = { ...input };
-  for (const key of INTENT_FIELDS) if (key in out && out[key] == null) delete out[key];
-  return out as T;
+  let out: Record<string, unknown> | undefined;
+  for (const key of INTENT_FIELDS) {
+    if (!(key in input)) continue;
+    const value = normalizeIntentField(key, input[key]);
+    if (value === input[key]) continue;
+    out ??= { ...input };
+    if (value === undefined) delete out[key];
+    else out[key] = value;
+  }
+  return (out ?? args) as T;
 }
 /**
- * Validate structured intent fields. Absent fields (omitted, undefined, or null) are fine; malformed
+ * Validate structured intent fields after `withoutAbsentIntent`. Absent fields are fine; malformed
  * ones are an error, and the caller must not run the command. `names` renames fields in the error
- * (e.g. snake_case parameters).
+ * (e.g. snake_case parameters). `0` in `expectedExitCodes` is accepted and dropped (#332).
  */
 export function readCommandIntent(args: unknown, names: Partial<Record<CommandIntentField, string>> = {}): { intent: CommandIntent; error?: string } {
   const input = withoutAbsentIntent((args && typeof args === "object" ? args : {}) as Record<string, unknown>);
@@ -138,7 +174,7 @@ export function readCommandIntent(args: unknown, names: Partial<Record<CommandIn
   if (codes !== undefined) {
     if (!Array.isArray(codes) || codes.length === 0 || codes.length > MAX_EXPECTED_EXIT_CODES ||
       !codes.every((code) => Number.isInteger(code) && code >= 1 && code <= 255) || new Set(codes).size !== codes.length) {
-      return { intent: {}, error: `${names.expectedExitCodes ?? "expectedExitCodes"} must be 1-${MAX_EXPECTED_EXIT_CODES} distinct integers from 1 to 255` };
+      return { intent: {}, error: `${names.expectedExitCodes ?? "expectedExitCodes"} must be 1-${MAX_EXPECTED_EXIT_CODES} distinct integers from 1 to 255 (0 is allowed and ignored)` };
     }
     intent.expectedExitCodes = [...codes] as number[];
   }

package/node_modules/pi-better-background-tasks/src/shared-log-utils.ts

@@ -819,29 +819,22 @@ export interface OutputControls {
   maxBytes?: unknown;
   /** Raw requested line count (canonical `lines`, else deprecated `tail_lines`). */
   lines?: unknown;
-  /** Deprecated alias names the caller used (whether or not the canonical name won). */
-  deprecated: string[];
 }
 
 /**
  * Resolve output-control parameters. Both spellings work; when both are given
- * the canonical name wins. Values are returned unvalidated so each surface
- * keeps its own defaults and hard caps.
+ * the canonical name wins. An explicit `null` is "not given", so a null
+ * canonical value never hides an alias value (#332). Values are returned
+ * unvalidated so each surface keeps its own defaults and hard caps.
  */
 export function readOutputControls(params: unknown): OutputControls {
   const p = (params && typeof params === "object" ? params : {}) as Record<string, unknown>;
-  const deprecated: string[] = [];
-  const pick = (canonical: keyof typeof OUTPUT_CONTROL_ALIASES): unknown => {
-    const alias = OUTPUT_CONTROL_ALIASES[canonical];
-    if (p[alias] !== undefined) deprecated.push(alias);
-    return p[canonical] !== undefined ? p[canonical] : p[alias];
-  };
+  const pick = (canonical: keyof typeof OUTPUT_CONTROL_ALIASES): unknown => p[canonical] ?? p[OUTPUT_CONTROL_ALIASES[canonical]] ?? undefined;
   const maxBytes = pick("max_bytes");
   const lines = pick("lines");
   return {
     ...(maxBytes !== undefined ? { maxBytes } : {}),
     ...(lines !== undefined ? { lines } : {}),
-    deprecated,
   };
 }
 
@@ -851,7 +844,8 @@ export type OutputInclude = (typeof OUTPUT_INCLUDE_VALUES)[number];
 
 /** Parse an `include` request into known values and unknown ones (both deduplicated). */
 export function readOutputInclude(value: unknown): { include: Set<OutputInclude>; unknown: string[] } {
-  const raw = Array.isArray(value) ? value : typeof value === "string" ? value.split(",") : [];
+  // Any other non-null value (a number, boolean, or object) is one unknown entry, never silently nothing (#332).
+  const raw = Array.isArray(value) ? value : typeof value === "string" ? value.split(",") : value == null ? [] : [value];
   const include = new Set<OutputInclude>();
   const unknown: string[] = [];
   for (const entry of raw) {